Make-Up FLOW: A Beauty YouTubers’ Video Recommendation Method Based on Make-Up Flowcharts

Make-up FLOW: A Beauty YouTubers’ Video Recommendation Method Based on Make-up Flowcharts Sayaka Takano and Satoshi Nakamura Meiji University, Japan 1

Background Diverse and complex makeup processes • Many people wear makeup depending on the situation for that day such as time, place, occasion and mood. Average maximum number of makeup processes: 16.0 Average number of makeup routes: 5.7 (by 34 people) • When applying makeup in their way, many people lack confidence in their makeup skills. 77% are self-taught in makeup. (by 984 people) [Kajita et al, ’22] Miho Kajita, Satoshi Nakamura. Basic Research on How to Apply Foundation Makeup Evenly on Your Own Face, 20th IFIP TC14 International Conference on Entertainment Computing (IFIP ICEC 2021), pp.402-410, 2021. 2

Background Increasing popularity of beauty information on SNS • 92% cosmetics consumers rely on reviews. • 57% of them consider SNS information crucial for purchasing new cosmetics. (by 11,115 people) [PowerReviews, ’22] • Viewers’ trust in beauty YouTubers significantly influences their willingness to buy products. [Chen et al, ’20] PowerReviews. https://www.powerreviews.com/research/beauty-shopper-digital-expectati ons-2022/. Accessed 7 Apr 2024 Chen, J., Dermawan, A.: The Influence of YouTube Beauty Vloggers on Indonesian Consumers’ Purchase Intention of Local Cosmeti c Products. International Journal of Business and Management, 15(5), 100-116 (2020) 3

Background Finding a suitable makeup tutorial is challenging. • There are 65 million makeup tutorials on the Internet. • Makeup involves many processes. • Many people have unique methods of makeup. 4

Base makeup A B Eyebrow makeup Eye makeup Shade makeup Lip makeup 5

Purpose To support updates to the makeup process by recommending makeup tutorials based on the makeup similarities Step1 Step2 Step3 Implementation of a system of makeup flowcharting Examination of a method for calculating the similarity of makeup processes Conducting an experiment to recommend makeup videos 6

Propose Make-up FLOW 7

Propose Make-up FLOW Face part Cosmetic item Eyeshadow Powder Item’s texture 4 branching conditions • Presence or absence of motivation • Difference in season • Length of time out of the house • Length of time spent on makeup 8

Propose Make-up FLOW This is a video of creating a flowchart of my makeup process using Makeup FLOW. 9

Construct a beauty YouTubers dataset Participants created flowcharts for 2 genres of makeup tutorial videos posted in the past year. 1. Everyday makeup (query: Everyday, Time reduction, …) 2. Special makeup (query: Look good, Popular, Party, …) We obtained 103 makeup flowcharts from 53 beauty YouTubers. (53 everyday makeup + 50 special makeup) 10

Construct a beauty YouTubers dataset Participants created flowcharts for 2 genres of makeup tutorial videos posted in the past year. 1. Everyday makeup (query: Everyday, Time reduction, …) 2. Special makeup (query: Look good, Popular, Party, …) University students Beauty YouTuber Min Max Ave. Stdv. 5 29 16.0 5.7 9 42 19.5 6.5 11

Base makeup Eyebrow makeup Eye makeup Shade makeup Lip makeup 12

Method for calculating the process similarity A Makeup base A Makeup base B Foundation (liquid) B Foundation (liquid) E Lip C1 Eyebrow(mascara) C Eyebrow(pencil) C Eyebrow(pencil) D Eyeshadow(powder) D Eyeshadow(powder) 𝐷1 Eyeshadow(liquid) E Lip E Lip G Cheek(powder) A Makeup base AB𝐶1 DE High ABCDE Low AEC𝐷1 G 13

Experiment to recommend makeup tutorials Recommend videos based on the makeup similarity between the user and a beauty YouTuber’s video. Participants watch and evaluate eight makeup videos. • Two high similarity videos and two average similarity videos by the standardized Levenshtein distance • Two high similarity videos and two average similarity videos by the cosine similarity based on N-gram frequency 14

Experiment to recommend makeup tutorials We calculated Spearman’s rank correlation coefficient between the similarity of the makeup process and the video ratings. video1 video2 Evaluation rank 1st 2nd Levenshtein rank 12th 56th N-gram rank 73th 43th … … … … video8 8th 103th 89th Spearman’s correlation coefficient 15

Results: Correlation coefficient Entire makeup process Levenshtein 0.26 N-gram 0.09 No correlation Base makeup Makeup to create a mood 16

Results: Correlation coefficient First / second makeup process 1st half 2nd half Levenshtein 0.38 0.21 N-gram 0.17 0.02 Maximum in the first half of the makeup process Levenshtein A 0.32 B 0.64 C -0.07 D 0.12 E 0.88 N-gram 0.49 0.31 -0.18 0.07 0.16 17

Results: Correlation coefficient Correlation coefficient for each participant It was helpful to know how to apply base makeup, eye makeup, etc. Levenshtein A 0.32 B 0.64 C -0.07 D 0.12 E 0.88 N-gram 0.49 0.31 -0.18 0.07 0.16 18

Results: Correlation coefficient Correlation coefficient for each participant There was no new knowledge. Her original face was too good. Levenshtein A 0.32 B 0.64 C -0.07 D 0.12 E 0.88 N-gram 0.49 0.31 -0.18 0.07 0.16 19

Results: Questionnaire Comments for low-rated videos • Discussed the topics viewers already knew, or that were unsuitable for them Comments for high-rated videos • Described items viewers had never used before • Explained tips for effective makeup application in great detail 20

Discussion • Trust in beauty influencers is essential when watching their videos. [Ding et al, 2019] • Popular beauty YouTubers used professional lighting and sound. [Rasmussen, 2018] Various factors are relevant to the evaluation of the video. Ding, W., Henninger, C.E., Blazquez, M., Boardman, R.: Effects of beauty vloggers’ eWOM and sponsored advertising on Weibo. Social Commerce, pp. 235-253 (2019). Rasmussen, L.L.: Parasocial interaction in the digital age: an examination of relationship building and the effectiveness of YouTube celebrities. Soc. Media Soc., 7(1), 280-294 (2018). 21

Future work Support for incorporating other’s makeup processes by sharing makeup flowcharts. 22

Summary Background Finding a suitable makeup video on SNS is challenging. Purpose Realization of video recommendation service based on similarity of makeup processes. Method Flowchart the makeup processes and recommend videos based on similarity of them. Result Recommendations based solely on similarity of makeup processes are inaccurate. 23